experiment_manifest.json

{
  "experiments": [
    {
      "id": "01_dispatch_basics",
      "display_name": "Dispatch Basics",
      "category": "foundations",
      "adapter_symbol": "run_dispatch_basics_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/dispatch_basics_experiment.cpp",
      "adapter_source": "src/experiments/adapters/dispatch_basics_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_dispatch_basics.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": false,
      "docs_group": "core",
      "plan_title": "Dispatch Basics",
      "plan_path": "experiments/01_dispatch_basics/plan.md",
      "plan_description": "Minimal Vulkan compute dispatch, correctness path, and baseline GPU timing."
    },
    {
      "id": "02_local_size_sweep",
      "display_name": "Local Size Sweep",
      "category": "foundations",
      "adapter_symbol": "run_local_size_sweep_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/local_size_sweep_experiment.cpp",
      "adapter_source": "src/experiments/adapters/local_size_sweep_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_local_size_sweep.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": false,
      "docs_group": "core",
      "plan_title": "Local Size Sweep",
      "plan_path": "experiments/02_local_size_sweep/plan.md",
      "plan_description": "Workgroup sizing and execution efficiency tradeoffs."
    },
    {
      "id": "03_memory_copy_baseline",
      "display_name": "Memory Copy Baseline",
      "category": "foundations",
      "adapter_symbol": "run_memory_copy_baseline_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/memory_copy_baseline_experiment.cpp",
      "adapter_source": "src/experiments/adapters/memory_copy_baseline_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_memory_copy_baseline.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": false,
      "docs_group": "core",
      "plan_title": "Memory Copy Baseline",
      "plan_path": "experiments/03_memory_copy_baseline/plan.md",
      "plan_description": "Raw buffer read/write/copy throughput characterization."
    },
    {
      "id": "04_sequential_indexing",
      "display_name": "Sequential Indexing",
      "category": "foundations",
      "adapter_symbol": "run_sequential_indexing_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/sequential_indexing_experiment.cpp",
      "adapter_source": "src/experiments/adapters/sequential_indexing_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_sequential_indexing.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": false,
      "docs_group": "core",
      "plan_title": "Sequential Indexing",
      "plan_path": "experiments/04_sequential_indexing/plan.md",
      "plan_description": "Ideal contiguous thread-to-data mapping as a good-path baseline."
    },
    {
      "id": "05_global_id_mapping_variants",
      "display_name": "Global ID Mapping Variants",
      "category": "foundations",
      "adapter_symbol": "run_global_id_mapping_variants_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/global_id_mapping_variants_experiment.cpp",
      "adapter_source": "src/experiments/adapters/global_id_mapping_variants_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_global_id_mapping_variants.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": false,
      "docs_group": "core",
      "plan_title": "Global ID Mapping Variants",
      "plan_path": "experiments/05_global_id_mapping_variants/plan.md",
      "plan_description": "Direct, offset, and grid-stride mapping behavior."
    },
    {
      "id": "06_aos_vs_soa",
      "display_name": "AoS vs SoA",
      "category": "memory_layout",
      "adapter_symbol": "run_aos_soa_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/aos_soa_experiment.cpp",
      "adapter_source": "src/experiments/adapters/aos_soa_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_aos_vs_soa.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": false,
      "docs_group": "core",
      "plan_title": "AoS vs SoA",
      "plan_path": "experiments/06_aos_vs_soa/plan.md",
      "plan_description": "Array-of-Structures versus Structure-of-Arrays layout efficiency.",
      "memory_sequence_order": 1
    },
    {
      "id": "07_aosoa_blocked_layout",
      "display_name": "AoSoA Blocked Layout",
      "category": "memory_layout",
      "adapter_symbol": "run_aosoa_blocked_layout_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/aosoa_blocked_layout_experiment.cpp",
      "adapter_source": "src/experiments/adapters/aosoa_blocked_layout_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_aosoa_blocked_layout.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": false,
      "docs_group": "core",
      "plan_title": "AoSoA or Blocked Layout",
      "plan_path": "experiments/07_aosoa_blocked_layout/plan.md",
      "plan_description": "Hybrid layout balancing vector locality and contiguous field access."
    },
    {
      "id": "08_std430_std140_packed",
      "display_name": "std430 vs std140 vs Packed",
      "category": "memory_layout",
      "adapter_symbol": "run_std430_std140_packed_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/std430_std140_packed_experiment.cpp",
      "adapter_source": "src/experiments/adapters/std430_std140_packed_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_std430_std140_packed.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": false,
      "docs_group": "core",
      "plan_title": "std430 vs std140 vs Packed",
      "plan_path": "experiments/08_std430_std140_packed/plan.md",
      "plan_description": "Shader buffer layout standards and padding cost."
    },
    {
      "id": "09_vec3_vec4_padding_costs",
      "display_name": "vec3 vs vec4 Padding Costs",
      "category": "memory_layout",
      "adapter_symbol": "run_vec3_vec4_padding_costs_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/vec3_vec4_padding_costs_experiment.cpp",
      "adapter_source": "src/experiments/adapters/vec3_vec4_padding_costs_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_vec3_vec4_padding_costs.py",
      "has_plot_script": false,
      "supports_skip_current": true,
      "plot_after_skip_current": false,
      "docs_group": "core",
      "plan_title": "vec3, vec4, and Padding Costs",
      "plan_path": "experiments/09_vec3_vec4_padding_costs/plan.md",
      "plan_description": "Impact of vector shape choice on storage efficiency and bandwidth."
    },
    {
      "id": "10_scalar_type_width_sweep",
      "display_name": "Scalar Type Width Sweep",
      "category": "memory_layout",
      "adapter_symbol": "run_scalar_type_width_sweep_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/scalar_type_width_sweep_experiment.cpp",
      "adapter_source": "src/experiments/adapters/scalar_type_width_sweep_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_scalar_type_width_sweep.py",
      "has_plot_script": false,
      "supports_skip_current": false,
      "plot_after_skip_current": false,
      "docs_group": "core",
      "plan_title": "Scalar Type Width Sweep",
      "plan_path": "experiments/10_scalar_type_width_sweep/plan.md",
      "plan_description": "Precision-width tradeoffs: 32-bit, 16-bit, and narrower storage."
    },
    {
      "id": "11_coalesced_vs_strided",
      "display_name": "Coalesced vs Strided Access",
      "category": "access_patterns",
      "adapter_symbol": "run_coalesced_vs_strided_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/coalesced_vs_strided_experiment.cpp",
      "adapter_source": "src/experiments/adapters/coalesced_vs_strided_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_coalesced_vs_strided.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Coalesced vs Strided Access",
      "plan_path": "experiments/11_coalesced_vs_strided/plan.md",
      "plan_description": "Contiguous and strided load behavior.",
      "memory_sequence_order": 2
    },
    {
      "id": "12_gather_access_pattern",
      "display_name": "Gather Access Pattern",
      "category": "access_patterns",
      "adapter_symbol": "run_gather_access_pattern_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/gather_access_pattern_experiment.cpp",
      "adapter_source": "src/experiments/adapters/gather_access_pattern_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_gather_access_pattern.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Gather Access Pattern",
      "plan_path": "experiments/12_gather_access_pattern/plan.md",
      "plan_description": "Indirect indexed reads through an index buffer."
    },
    {
      "id": "13_scatter_access_pattern",
      "display_name": "Scatter Access Pattern",
      "category": "access_patterns",
      "adapter_symbol": "run_scatter_access_pattern_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/scatter_access_pattern_experiment.cpp",
      "adapter_source": "src/experiments/adapters/scatter_access_pattern_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_scatter_access_pattern.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Scatter Access Pattern",
      "plan_path": "experiments/13_scatter_access_pattern/plan.md",
      "plan_description": "Indirect indexed writes and contention behavior."
    },
    {
      "id": "14_read_reuse_cache_locality",
      "display_name": "Read Reuse and Cache Locality",
      "category": "access_patterns",
      "adapter_symbol": "run_read_reuse_cache_locality_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/read_reuse_cache_locality_experiment.cpp",
      "adapter_source": "src/experiments/adapters/read_reuse_cache_locality_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_read_reuse_cache_locality.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Read Reuse and Cache Locality",
      "plan_path": "experiments/14_read_reuse_cache_locality/plan.md",
      "plan_description": "Temporal locality and reuse-distance effects.",
      "memory_sequence_order": 3
    },
    {
      "id": "15_bandwidth_saturation_sweep",
      "display_name": "Bandwidth Saturation Sweep",
      "category": "access_patterns",
      "adapter_symbol": "run_bandwidth_saturation_sweep_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/bandwidth_saturation_sweep_experiment.cpp",
      "adapter_source": "src/experiments/adapters/bandwidth_saturation_sweep_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_bandwidth_saturation_sweep.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Bandwidth Saturation Sweep",
      "plan_path": "experiments/15_bandwidth_saturation_sweep/plan.md",
      "plan_description": "Scaling data volume until practical bandwidth plateau.",
      "memory_sequence_order": 4
    },
    {
      "id": "16_shared_memory_tiling",
      "display_name": "Shared Memory Tiling",
      "category": "on_chip_memory",
      "adapter_symbol": "run_shared_memory_tiling_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/shared_memory_tiling_experiment.cpp",
      "adapter_source": "src/experiments/adapters/shared_memory_tiling_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_shared_memory_tiling.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Shared or Workgroup Memory Tiling",
      "plan_path": "experiments/16_shared_memory_tiling/plan.md",
      "plan_description": "Staging data in on-chip memory for reuse.",
      "memory_sequence_order": 5
    },
    {
      "id": "17_tile_size_sweep",
      "display_name": "Tile Size Sweep",
      "category": "on_chip_memory",
      "adapter_symbol": "run_tile_size_sweep_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/tile_size_sweep_experiment.cpp",
      "adapter_source": "src/experiments/adapters/tile_size_sweep_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_tile_size_sweep.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Tile Size Sweep",
      "plan_path": "experiments/17_tile_size_sweep/plan.md",
      "plan_description": "Tradeoff between reuse, shared-memory pressure, and occupancy."
    },
    {
      "id": "18_register_pressure_proxy",
      "display_name": "Register Pressure Proxy",
      "category": "execution_model",
      "adapter_symbol": "run_register_pressure_proxy_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/register_pressure_proxy_experiment.cpp",
      "adapter_source": "src/experiments/adapters/register_pressure_proxy_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_register_pressure_proxy.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Register Pressure Proxy Study",
      "plan_path": "experiments/18_register_pressure_proxy/plan.md",
      "plan_description": "Effect of increased per-thread temporary state."
    },
    {
      "id": "19_branch_divergence",
      "display_name": "Branch Divergence",
      "category": "execution_model",
      "adapter_symbol": "run_branch_divergence_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/branch_divergence_experiment.cpp",
      "adapter_source": "src/experiments/adapters/branch_divergence_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_branch_divergence.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Branch Divergence",
      "plan_path": "experiments/19_branch_divergence/plan.md",
      "plan_description": "Control-flow divergence within warp or wave execution."
    },
    {
      "id": "20_barrier_synchronization_cost",
      "display_name": "Barrier and Synchronization Cost",
      "category": "on_chip_memory",
      "adapter_symbol": "run_barrier_synchronization_cost_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/barrier_synchronization_cost_experiment.cpp",
      "adapter_source": "src/experiments/adapters/barrier_synchronization_cost_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_barrier_synchronization_cost.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Barrier and Synchronization Cost",
      "plan_path": "experiments/20_barrier_synchronization_cost/plan.md",
      "plan_description": "Synchronization overhead characterization."
    },
    {
      "id": "21_parallel_reduction",
      "display_name": "Parallel Reduction",
      "category": "parallel_primitives",
      "adapter_symbol": "run_parallel_reduction_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/parallel_reduction_experiment.cpp",
      "adapter_source": "src/experiments/adapters/parallel_reduction_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_parallel_reduction.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Parallel Reduction",
      "plan_path": "experiments/21_parallel_reduction/plan.md",
      "plan_description": "Reduction patterns from naive to tree and shared-memory optimized."
    },
    {
      "id": "22_prefix_sum_scan",
      "display_name": "Prefix Sum Scan",
      "category": "parallel_primitives",
      "adapter_symbol": "run_prefix_sum_scan_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/prefix_sum_scan_experiment.cpp",
      "adapter_source": "src/experiments/adapters/prefix_sum_scan_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_prefix_sum_scan.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Prefix Sum or Scan",
      "plan_path": "experiments/22_prefix_sum_scan/plan.md",
      "plan_description": "Inclusive/exclusive scan as a foundational parallel primitive."
    },
    {
      "id": "23_histogram_atomic_contention",
      "display_name": "Histogram and Atomic Contention",
      "category": "parallel_primitives",
      "adapter_symbol": "run_histogram_atomic_contention_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/histogram_atomic_contention_experiment.cpp",
      "adapter_source": "src/experiments/adapters/histogram_atomic_contention_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_histogram_atomic_contention.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Histogram and Atomic Contention",
      "plan_path": "experiments/23_histogram_atomic_contention/plan.md",
      "plan_description": "Atomic update contention and privatization strategies."
    },
    {
      "id": "24_stream_compaction",
      "display_name": "Stream Compaction",
      "category": "parallel_primitives",
      "adapter_symbol": "run_stream_compaction_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/stream_compaction_experiment.cpp",
      "adapter_source": "src/experiments/adapters/stream_compaction_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_stream_compaction.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Stream Compaction",
      "plan_path": "experiments/24_stream_compaction/plan.md",
      "plan_description": "Flag, scan, and compact-write pipeline."
    },
    {
      "id": "25_spatial_binning_clustered_culling_capstone",
      "display_name": "Spatial Binning or Clustered Culling Capstone",
      "category": "parallel_primitives",
      "adapter_symbol": "run_spatial_binning_clustered_culling_capstone_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/spatial_binning_clustered_culling_capstone_experiment.cpp",
      "adapter_source": "src/experiments/adapters/spatial_binning_clustered_culling_capstone_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_spatial_binning_clustered_culling_capstone.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "core",
      "plan_title": "Spatial Binning or Clustered Culling Capstone",
      "plan_path": "experiments/25_spatial_binning_clustered_culling_capstone/plan.md",
      "plan_description": "Rendering-style compute pipeline combining prior primitives."
    },
    {
      "id": "26_warp_level_coalescing_alignment",
      "display_name": "Warp-Level Coalescing Alignment",
      "category": "access_patterns",
      "adapter_symbol": "run_warp_level_coalescing_alignment_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/warp_level_coalescing_alignment_experiment.cpp",
      "adapter_source": "src/experiments/adapters/warp_level_coalescing_alignment_adapter.cpp",
      "default_size": "2G",
      "analysis_script": "analyze_warp_level_coalescing_alignment.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "extension",
      "plan_title": "Warp-Level Coalescing Alignment",
      "plan_path": "experiments/26_warp_level_coalescing_alignment/plan.md",
      "plan_description": "Aligned vs misaligned contiguous accesses at warp granularity."
    },
    {
      "id": "27_cache_thrashing_random_vs_sequential",
      "display_name": "Cache Thrashing, Random vs Sequential",
      "category": "access_patterns",
      "adapter_symbol": "run_cache_thrashing_random_vs_sequential_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/cache_thrashing_random_vs_sequential_experiment.cpp",
      "adapter_source": "src/experiments/adapters/cache_thrashing_random_vs_sequential_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_cache_thrashing_random_vs_sequential.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "extension",
      "plan_title": "Cache Thrashing, Random vs Sequential",
      "plan_path": "experiments/27_cache_thrashing_random_vs_sequential/plan.md",
      "plan_description": "Healthy locality versus deliberate cache defeat."
    },
    {
      "id": "28_device_local_vs_host_visible_heap_placement",
      "display_name": "Device-Local vs Host-Visible Heap Placement",
      "category": "memory_hierarchy",
      "adapter_symbol": "run_device_local_vs_host_visible_heap_placement_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/device_local_vs_host_visible_heap_placement_experiment.cpp",
      "adapter_source": "src/experiments/adapters/device_local_vs_host_visible_heap_placement_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_device_local_vs_host_visible_heap_placement.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "extension",
      "plan_title": "Device-Local vs Host-Visible Heap Placement",
      "plan_path": "experiments/28_device_local_vs_host_visible_heap_placement/plan.md",
      "plan_description": "Dispatch-only and end-to-end cost of host-visible buffers versus staged device-local placement."
    },
    {
      "id": "29_shared_memory_bank_conflict_study",
      "display_name": "Shared Memory Bank Conflict Study",
      "category": "on_chip_memory",
      "adapter_symbol": "run_shared_memory_bank_conflict_study_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/shared_memory_bank_conflict_study_experiment.cpp",
      "adapter_source": "src/experiments/adapters/shared_memory_bank_conflict_study_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_shared_memory_bank_conflict_study.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "extension",
      "plan_title": "Shared Memory Bank Conflict Study",
      "plan_path": "experiments/29_shared_memory_bank_conflict_study/plan.md",
      "plan_description": "Stride-driven shared-memory bank conflicts and the padding fix."
    },
    {
      "id": "30_subgroup_reduction_variants",
      "display_name": "Subgroup Reduction Variants",
      "category": "parallel_primitives",
      "adapter_symbol": "run_subgroup_reduction_variants_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/subgroup_reduction_variants_experiment.cpp",
      "adapter_source": "src/experiments/adapters/subgroup_reduction_variants_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_subgroup_reduction_variants.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "extension",
      "plan_title": "Subgroup Reduction Variants",
      "plan_path": "experiments/30_subgroup_reduction_variants/plan.md",
      "plan_description": "Compare shared-tree reduction with subgroup-assisted reduction."
    },
    {
      "id": "31_subgroup_scan_variants",
      "display_name": "Subgroup Scan Variants",
      "category": "parallel_primitives",
      "adapter_symbol": "run_subgroup_scan_variants_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/subgroup_scan_variants_experiment.cpp",
      "adapter_source": "src/experiments/adapters/subgroup_scan_variants_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_subgroup_scan_variants.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "extension",
      "plan_title": "Subgroup Scan Variants",
      "plan_path": "experiments/31_subgroup_scan_variants/plan.md",
      "plan_description": "Block-local inclusive scan using shared memory versus subgroup intrinsics."
    },
    {
      "id": "32_subgroup_stream_compaction_variants",
      "display_name": "Subgroup Stream Compaction Variants",
      "category": "parallel_primitives",
      "adapter_symbol": "run_subgroup_stream_compaction_variants_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/subgroup_stream_compaction_variants_experiment.cpp",
      "adapter_source": "src/experiments/adapters/subgroup_stream_compaction_variants_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_subgroup_stream_compaction_variants.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "extension",
      "plan_title": "Subgroup Stream Compaction Variants",
      "plan_path": "experiments/32_subgroup_stream_compaction_variants/plan.md",
      "plan_description": "Per-workgroup compaction using shared atomics versus subgroup ballot ranking."
    },
    {
      "id": "33_two_dimensional_locality_transpose_study",
      "display_name": "2D Locality and Transpose Study",
      "category": "access_patterns",
      "adapter_symbol": "run_two_dimensional_locality_transpose_study_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/two_dimensional_locality_transpose_study_experiment.cpp",
      "adapter_source": "src/experiments/adapters/two_dimensional_locality_transpose_study_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_two_dimensional_locality_transpose_study.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "extension",
      "plan_title": "2D Locality and Transpose Study",
      "plan_path": "experiments/33_two_dimensional_locality_transpose_study/plan.md",
      "plan_description": "Row-major copy versus naive and tiled transpose access patterns."
    },
    {
      "id": "34_radix_sort_gpu",
      "display_name": "Radix Sort on GPU",
      "category": "parallel_primitives",
      "adapter_symbol": "run_radix_sort_gpu_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/radix_sort_gpu_experiment.cpp",
      "adapter_source": "src/experiments/adapters/radix_sort_gpu_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_radix_sort_gpu.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "advanced",
      "plan_title": "Radix Sort on GPU",
      "plan_path": "experiments/34_radix_sort_gpu/plan.md",
      "plan_description": "Multi-pass GPU radix sort: 8-bit vs 4-bit digit width variants, key-only."
    },
    {
      "id": "35_occupancy_modeling",
      "display_name": "Occupancy Modeling",
      "category": "compute_throughput",
      "adapter_symbol": "run_occupancy_modeling_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/occupancy_modeling_experiment.cpp",
      "adapter_source": "src/experiments/adapters/occupancy_modeling_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_occupancy_modeling.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "advanced",
      "plan_title": "Occupancy Modeling Against Vendor Guidance",
      "plan_path": "experiments/35_occupancy_modeling/plan.md",
      "plan_description": "Low, medium, and high shared-memory pressure variants to model occupancy effects on throughput."
    },
    {
      "id": "36_bvh_node_layout",
      "display_name": "BVH Node Layout",
      "category": "rendering_systems",
      "adapter_symbol": "run_bvh_node_layout_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/bvh_node_layout_experiment.cpp",
      "adapter_source": "src/experiments/adapters/bvh_node_layout_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_bvh_node_layout.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "advanced",
      "plan_title": "BVH Node Layout",
      "plan_path": "experiments/36_bvh_node_layout/plan.md",
      "plan_description": "Compact versus padded BVH-node storage and hashed traversal-order locality impact."
    },
    {
      "id": "37_frame_to_frame_coherence",
      "display_name": "Frame-to-Frame Coherence",
      "category": "rendering_systems",
      "adapter_symbol": "run_frame_to_frame_coherence_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/frame_to_frame_coherence_experiment.cpp",
      "adapter_source": "src/experiments/adapters/frame_to_frame_coherence_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_frame_to_frame_coherence.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "advanced",
      "plan_title": "Frame-to-Frame Coherence",
      "plan_path": "experiments/37_frame_to_frame_coherence/plan.md",
      "plan_description": "Temporal locality sensitivity across coherent, block-scrambled, and frame-random access."
    },
    {
      "id": "38_frustum_vs_clustered_culling",
      "display_name": "Frustum vs Clustered Culling",
      "category": "rendering_systems",
      "adapter_symbol": "run_frustum_vs_clustered_culling_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/frustum_vs_clustered_culling_experiment.cpp",
      "adapter_source": "src/experiments/adapters/frustum_vs_clustered_culling_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_frustum_vs_clustered_culling.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "advanced",
      "plan_title": "Frustum Culling vs Clustered Culling",
      "plan_path": "experiments/38_frustum_vs_clustered_culling/plan.md",
      "plan_description": "Visibility list construction under direct frustum filtering versus clustered binning."
    },
    {
      "id": "39_tiled_light_assignment",
      "display_name": "Tiled Light Assignment",
      "category": "rendering_systems",
      "adapter_symbol": "run_tiled_light_assignment_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/tiled_light_assignment_experiment.cpp",
      "adapter_source": "src/experiments/adapters/tiled_light_assignment_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_tiled_light_assignment.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "advanced",
      "plan_title": "Tiled Light Assignment",
      "plan_path": "experiments/39_tiled_light_assignment/plan.md",
      "plan_description": "Compare light-centric atomics against tile-centric counting strategies."
    },
    {
      "id": "40_persistent_threads_work_queues",
      "display_name": "Persistent Threads and Work Queues",
      "category": "execution_model",
      "adapter_symbol": "run_persistent_threads_work_queues_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/persistent_threads_work_queues_experiment.cpp",
      "adapter_source": "src/experiments/adapters/persistent_threads_work_queues_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_persistent_threads_work_queues.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "advanced",
      "plan_title": "Persistent Threads and Work Queues",
      "plan_path": "experiments/40_persistent_threads_work_queues/plan.md",
      "plan_description": "Static partitioning versus persistent queue workers under irregular task costs."
    },
    {
      "id": "41_subgroup_operations_study",
      "display_name": "Subgroup Operations Study",
      "category": "parallel_primitives",
      "adapter_symbol": "run_subgroup_operations_study_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/subgroup_operations_study_experiment.cpp",
      "adapter_source": "src/experiments/adapters/subgroup_operations_study_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_subgroup_operations_study.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "advanced",
      "plan_title": "Subgroup Operations Study",
      "plan_path": "experiments/41_subgroup_operations_study/plan.md",
      "plan_description": "Shared-memory baseline versus subgroup arithmetic and ballot intrinsics."
    },
    {
      "id": "42_async_compute_overlap",
      "display_name": "Async Compute Overlap",
      "category": "compute_throughput",
      "adapter_symbol": "run_async_compute_overlap_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/async_compute_overlap_experiment.cpp",
      "adapter_source": "src/experiments/adapters/async_compute_overlap_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_async_compute_overlap.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "advanced",
      "plan_title": "Async Compute Overlap",
      "plan_path": "experiments/42_async_compute_overlap/plan.md",
      "plan_description": "Serial two-pass execution versus fused overlap proxy on independent workloads."
    },
    {
      "id": "43_ray_friendly_memory_layouts",
      "display_name": "Ray-Friendly Memory Layouts",
      "category": "rendering_systems",
      "adapter_symbol": "run_ray_friendly_memory_layouts_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/ray_friendly_memory_layouts_experiment.cpp",
      "adapter_source": "src/experiments/adapters/ray_friendly_memory_layouts_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_ray_friendly_memory_layouts.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "advanced",
      "plan_title": "Ray-Friendly Memory Layouts",
      "plan_path": "experiments/43_ray_friendly_memory_layouts/plan.md",
      "plan_description": "AoS-style padded and SoA-style compact layouts under sequential and hashed ray access."
    },
    {
      "id": "44_gpu_driven_pipeline_blocks",
      "display_name": "GPU-Driven Pipeline Blocks",
      "category": "rendering_systems",
      "adapter_symbol": "run_gpu_driven_pipeline_blocks_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/gpu_driven_pipeline_blocks_experiment.cpp",
      "adapter_source": "src/experiments/adapters/gpu_driven_pipeline_blocks_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_gpu_driven_pipeline_blocks.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "advanced",
      "plan_title": "GPU-Driven Pipeline Building Blocks",
      "plan_path": "experiments/44_gpu_driven_pipeline_blocks/plan.md",
      "plan_description": "Compares staged cull-bucket-emit dispatch pipelines against a fused single-dispatch implementation."
    },
    {
      "id": "45_cross_gpu_reproducibility",
      "display_name": "Cross-GPU Reproducibility",
      "category": "compute_throughput",
      "adapter_symbol": "run_cross_gpu_reproducibility_experiment_adapter",
      "enabled": true,
      "source": "src/experiments/cross_gpu_reproducibility_experiment.cpp",
      "adapter_source": "src/experiments/adapters/cross_gpu_reproducibility_adapter.cpp",
      "default_size": "1G",
      "analysis_script": "analyze_cross_gpu_reproducibility.py",
      "has_plot_script": true,
      "supports_skip_current": true,
      "plot_after_skip_current": true,
      "docs_group": "advanced",
      "plan_title": "Reproducibility and Cross-GPU Comparison",
      "plan_path": "experiments/45_cross_gpu_reproducibility/plan.md",
      "plan_description": "Deterministic coalesced, stride, and hashed probes with checksum notes for cross-run and cross-GPU comparison."
    }
  ]
}